The Study of Density-based Classification through Hyper-spectrum Image Data of Paddy Rice with Considering Attribute Reduction

以密度為基礎的演算法在高光譜水稻田影像的應用：考慮影像屬性刪減

萬絢(Shiuan Wan) ； 王依蘋(Yi-Ping Wang) ； 鄭育欣(Yu-Hsin Cheng)

《航測及遙測學刊》 25卷3期 (2020/09) Pp. 181-192

https://doi.org/10.6574/JPRS.202009_25(3).0004

摘要

In the past, it draws the great attraction of using Data Mining approaches and geostatistics analysis through remote sensing data which are well-accepted. The goal of this study is decided to extract the core spectral information through hyperspectral vs. multispectral imaging. More specifically, the paddy-field remote sensing image is applied with a supervised learning linear discriminant analysis and unsupervised learning density-based clustering algorithm in this study. The pre-processing is used the Principal Component Analysis (PCA) to design parallel study for four case studies: (1) hyper-spectrum versus multi-spectrum with linear discriminant analysis (2) hyper-spectrum versus multi-spectrum density-based clustering algorithm (3) hyper-spectrum versus multispectrum principal component analysis + linear discriminant analysis (4) hyper-spectrum versus multi-spectrum by principal component + density-based clustering algorithm. The DBSCAN with hyper-spectrum image data has an overall accuracy rate of 86.85% which is higher than those of DBSCAN with multi-spectrum (79.45%). The results are presented by error matrix (accuracy rate) and the thematic maps are drawn.

關鍵字

Image Classification ； Data Mining ； Density-based Clustering Algorithm

並列摘要

過去，資料挖掘方法和通過遙感數據吸引了人們的廣泛興趣且廣為接收。本研究的目的是通過高光 譜與多光譜成像提取有影響性之信息，本研究將稻田遙感圖像與監督學習線性判別分析和基於監督的學習密度的聚類算法一起應用。以密度為基礎的演算法在高光譜材料上有86.85%的辨識成果比傳統光譜79.45%高出很多。本研究使用主成分分析 （PCA）四個案例研究：（1）基於線性判別分析的高光譜與多光譜；（2）高光譜與多光譜基於密度的聚類算法；（3）高光譜與多光譜PCA+線性判別分析；（4）高光譜與多光譜 PCA+基於密度的聚類算法，結果由誤差矩陣（準確率）表示，並繪製了主題圖。

並列關鍵字

影像分類 ； 資料探勘 ； 基於密度的聚類算法